Frontier AI labs claim their models disproved a long-standing computer science assumption

If confirmed, the claimed algorithms would call into question research that rests on the hypothesis, while the community weighs how AI-generated results should be checked.

Frontier AI labs OpenAI and Anthropic have each claimed their models produced algorithms that solve a class of hard logic puzzles in sub-exponential time, which if confirmed would refute the Exponential Time Hypothesis, a long-standing assumption in computer science. The claim matters because the hypothesis supports the widespread belief that P does not equal NP, so a counterexample would shift how researchers think about which problems are computationally tractable. MIT's Ryan Williams is helping organise the computer science community's response, with an open letter seeking feedback and debate over norms for AI contributions. Williams also notes the work could affect many published papers that rely on the hypothesis, and that the labs checked whether any cryptographic systems would be broken and reportedly found none. He cautions that the algorithm is unlikely to be the final word.